Quantcast

Maximum PC

It is currently Fri Sep 19, 2014 12:35 am

All times are UTC - 8 hours




Post new topic Reply to topic  [ 5 posts ] 
Author Message
 Post subject: SSD speed this... SSD speed that...
PostPosted: Sat Apr 26, 2014 10:16 pm 
Smithfield
Smithfield

Joined: Sun Jun 18, 2006 7:37 pm
Posts: 5251
One thing that's sort of bothering me on what people are gushing about SSD speeds on is... how fast they can transfer. While that is an important factor, what really should be focused on is how fast they can get the data. An SSD running at 100MB/s and 0.1ms access time I'm sure is a lot more pleasant than an SSD running at 500MB/s with 20ms access time. For comparison sake, if I were to make 10,000 requests, the 0.1ms latency adds 1 second of overhead. 20ms of latency adds 200 seconds.

Or to put it in another perspective, say if we manage to get an SSD that saturates a PCI Express 3.0 x16 slot. Theoretically this amounts to 16GB/s of bandwidth, which is the equivalent of DDR3-2000. Sweet, now we have NVRAM!... until you realize that the latency on an SSD is still, give or take 10,000,000 slower than RAM (not to mention using an SSD as NVRAM will destroy it).

Also I might be slow on latency. That knowledge is about two years old.


Top
  Profile  
 
 Post subject: Re: SSD speed this... SSD speed that...
PostPosted: Tue May 06, 2014 1:59 pm 
Team Member Top 500
Team Member Top 500
User avatar

Joined: Thu Jul 21, 2005 10:45 pm
Posts: 3026
Location: Central Florida
But you're missing two very important points...
1> Latency is typically measured in ns(1,000,000,000/th second) or μs(1,000,000/th second) not ms(1,000/th second), - 1000μs = 1ms
2> Access Time is not Latency - though they are inseparably tied together Access Time = latency + the time required to reach the access point for the requested data and be ready to access or write the data.

Also, when people look at SSD speed, it's most often compared to HDD speeds.

Show me any decent SATA III SSD with an access time greater than 0.2ms... sure some of the cheap crap ones, but any of the decent ones from q4 2012 on up are all sub 0.2ms read & write access times with most of those being under 0.1ms. Even the best 15,000rpm enterprise HDD's have access times no faster than about 5.6ms & most are over 6ms. You have to drop down to SATA II SSD's before you see access times higher than 0.5ms.

For reference my two 256GB SATA III SSD's & my 2TB HDD (as recorded through AS-SSD)
OCZ Vector: 0.097ms read / 0.053ms write - 512MB/s read / 492MB/s write (desktop OS drive - & 57x lower read latency than any enterprise HDD)
Samsung 840 Pro: 0.064ms read / 0.030ms write - 517MB/s read / 360MB/s write (laptop OS drive)
WD Red NAS 7,200rpm: 20.292ms read / 13.608ms write 137MB/s read / 124MB/s write (primary desktop storage drive)

EDITED because wrong abbreviation was used & completely negated the second part of this post which has now been deleted.


Last edited by chaosdsm on Wed May 07, 2014 3:44 pm, edited 1 time in total.

Top
  Profile  
 
 Post subject: Re: SSD speed this... SSD speed that...
PostPosted: Tue May 06, 2014 2:43 pm 
Smithfield
Smithfield

Joined: Sun Jun 18, 2006 7:37 pm
Posts: 5251
Like I said, my knowledge on SSD latency was over two years old, because for all I care, they're bitchin' fast. :P

The only problem with SSDs is they're the only thing I'm aware of that gets progressively worse as we're trying to "improve it". I see SSD technology hitting the wall very soon. The only hope SSDs have is stacking memory before the node size and number of bits each cell can hold lowers the lifespan to hundreds or even tens.

Well, here's hoping memristors finally take off.


Top
  Profile  
 
 Post subject: Re: SSD speed this... SSD speed that...
PostPosted: Wed May 07, 2014 3:39 pm 
Team Member Top 500
Team Member Top 500
User avatar

Joined: Thu Jul 21, 2005 10:45 pm
Posts: 3026
Location: Central Florida
and yet, when compared to previous generations, every new generation of SSD has:
improved capacity
improved size to capacity ratio (i.e. even mSATA SSD's are now available in 1TB capacity)
reduced latency
improved data throughput speeds

While SLC > MLC > TLC has effectively reduced the number of Program/Erase cycles, increasingly lower $/capacity negates this effect...

For example, current TLC is rated at about 1,000 - 1,500 P/E cycles, using the lower 1,000 rate, and given a 256GB TLC SSD, you can continuously write 70GB of data, every day, 24/7/365, for 10 straight years before reaching 1,000 P/E cycles. To put it another way, in two years of use, my mom's 128GB SSD saw just 2.8TB of total writes (per S.M.A.R.T. data), given the same 256GB TLC SSD, that's 178 YEARS worth of use before exhausting the theoretical 1,000 P/E cycle limit.

IMO, there's currently only three areas of performance that needs improvement in SSD's, & all three relate to small file read/write. I want to see 4kb read/write performance over the 200MB/s mark & ideally at over half the maximum throughput speed of the SSD. But there are some serious obstacles, primarily latency & the speed at which the controller can handle multiple read/write requests. But we would need a controller that processes requests at least 10x faster than current controllers, and we would also need to see at least a 10x reduction in latency. Very tall orders on both counts.

Going from one 120GB SATA II SSD @ 200MB/s 0.36ms access-time, to one 120GB SATAIII SSD @ 380MB/s 0.22ms access time, to dual 128GB SATA III SSD's in RAID-0 @ 900MB/s 0.16ms access-time, & down to one 256GB SATA III SSD @ 500MB/s 0.09ms access-time, I can honestly say that the performance variance between all 4 setups was only noticeable with a stop-watch. That's primarily because, the majority of reads are small files. Changes between the three setups in 4k read benchmarks was tiny with the worst hitting 19.36MB/s read & the best hitting just 28.9MB/s read - not counting the 840 pro in "rapid" mode.

Doesn't matter if we move to SATA4 & get a SSD that can hit 4GB/s read/write speeds on the interface, if the small file transfer rates remain stagnant, there won't be much in the way of noticeable improvements in actual use.

Speaking of latency, I used the wrong abbreviation last time, μs should actually have been ns... current DDR3 RAM ranges from 15ns to 60ns latency http://www.memorybenchmark.net/latency_ddr3_intel.html by comparison the best SSD's have a whopping 30,000ns of latency, so SSD's are still WAY FAR BEHIND RAM.
ns = 1,000,000,000/th of a second
μs = 1,000,000/th of a second
ms = 1,000/th of a second

Actually, the white-paper I was reading used the wrong abbreviation...


Top
  Profile  
 
 Post subject: Re: SSD speed this... SSD speed that...
PostPosted: Wed May 07, 2014 3:54 pm 
Smithfield
Smithfield

Joined: Sun Jun 18, 2006 7:37 pm
Posts: 5251
Well, at least you see where the issue lies and where I see the value in faster memory. :3

I was also wondering where you got your DRAM statistics from, because I was pretty certain the cycle time in RAM was less than a nanosecond and total refresh cycles were still within 50 or less.


Top
  Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 5 posts ] 

All times are UTC - 8 hours


Who is online

Users browsing this forum: No registered users and 4 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group